← Back to Glossary
Alignment
Steering an LLM's behavior so its actions, tone, and safety constraints match human values and intentions.
Why it exists
Unchecked base models can generate biased or unhelpful content. Alignment ensures models remain safe, polite, and instruction-following.
How it works
Combines Supervised Fine-Tuning (SFT), RLHF/RLAIF, red-teaming, and rule-based safety guardrails.
Related terms
Where you'll use this
roadmap